This was a write up and article that I came across today that reminded me of what you were saying here about keeping users and coders in a flow state. Thought you may enjoy it if you hadn’t seen it already. 🙂

x.com/math_rachel/...

Looking forward to your new LLM post you mentioned in another reply!

Without that learning feedback, I find it disorienting. I’m not understanding the problem when an AI just spits out code.

Then when I dig in and do understand the problem, I find the AI didn’t understand the problem and just tackled the problem tactically without any strategic consideration.

So well said. My problem is that typing and thinking with the code tests my understanding of the problem and my model of it. That feedback with errors and mistakes is a critical signal that lets me know if I’m on the right track.

Once again, the problem is that people ignore all of the invisible costs of vibe coding:

- prompting (specifying what to do *takes time*)
- learned helplessness
- constantly interrupting their own flow state
- babysitting rules files
- reviewing the LLM's output
- fixing bugs introduced by the LLM